【Spring 2021】Independent Project – Automated Parameter Tuning for Land Ice Simulations
Global mean sea-level is rising at a rate of 3.2 mm/year and this rate is increasing, with the latest studies suggesting possible increase in sea-level of 0.3-2.5 m by 2100, due to melting of the polar ice sheets (Greenland, Antarctica). With this in mind, Sandia National Laboratories have been developing the MPAS-Albany Land Ice (MALI) Model, as part of the DOE’s Energy Exascale Earth System Model, to provide actionable projections of 21st century sea-level rise and support national security missions on high performance computing (HPC) systems.
Our team focused on building a framework for each type of data to facilitate the automated parameter tuning for ice sheet simulations of Earth’s polar ice sheets. This was part of a collaborative work of Sandia (lasted for two years, from 2019-2021) to modernize climate software and develop more accurate and reliable models for probabilistic sea-level projections, which is preprinted on arXiv in early 2022, and then further published on IJHPCA in 2023. In this work, we explored a subset of the input parameter space to tune performance, utilized grid and random search for optimization, and automated the tuning process with an efficient framework.
Relevant links:
- Link to the published journal on IJHPCA, The International Journal of High Performance Computing Applications
- Link to my working repository: Autotuning
- Link to my Final Report
- Link to our arXiv pre-print (cs>arXiv:2204.04321)